An antibody class with a common CDRH3 motif broadly neutralizes sarbecoviruses

The devastation caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has made clear the importance of pandemic preparedness. To address future zoonotic outbreaks due to related viruses in the sarbecovirus subgenus, we identified a human monoclonal antibody, 10-40, that neutralized or bound all sarbecoviruses tested in vitro and protected against SARS-CoV-2 and SARS-CoV in vivo. Comparative studies with other receptor-binding domain (RBD)-directed antibodies showed 10-40 to have the greatest breadth against sarbecoviruses, suggesting that 10-40 is a promising agent for pandemic preparedness. Moreover, structural analyses on 10-40 and similar antibodies not only defined an epitope cluster in the inner face of the RBD that is well-conserved among sarbecoviruses, but also uncovered a distinct antibody class with a common CDRH3 motif. Our analyses also suggested that elicitation of this class of antibodies may not be overly difficult, an observation that bodes well for the development of a pan-sarbecovirus vaccine.


INTRODUCTION
The coronavirus disease 2019 (COVID-19) pandemic is caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), which has infected more than 480 million people and resulted in more than 6.1 million deaths (1). Multiple variants of this virus have emerged, including some capable of increased transmission or antibody evasion (2)(3)(4). Furthermore, the threat of continued zoonotic spillovers warrants the development of interventions that could broadly combat animal coronaviruses with pandemic potential (5).
Numerous anti-SARS-CoV-2 monoclonal antibodies (mAbs) have been isolated and characterized, with several demonstrating clinical utility (6,7). Some have been reported to have broadly neutralizing activity against not only SARS-CoV-2 but also other sarbecoviruses (8)(9)(10)(11)(12)(13)(14)(15)(16), a viral subgenus containing both SARS-CoV-2 and SARS-CoV (17). Such broadly neutralizing mAbs could serve as a therapeutic adjunct for the current pandemic as well as a useful agent in addressing future zoonoses caused by sarbecoviruses. Given the therapeutic potential and vast array of mAbs that have been described to hold such breadth, a side-by-side evaluation of each of these mAbs is needed. Such a study would be valuable to understand the differing specificities of these mAbs and identify which would best serve as pandemic preparedness agents. Furthermore, identification of commonalities and deficiencies among mAbs would aid in designing antigens for use as a pan-sarbecovirus vaccine.
Here, we report the isolation of three mAbs that broadly neutralize sarbecoviruses. In addition to virological and structural studies of these three mAbs, we conduct a comprehensive comparative analysis of these mAbs together with nine other mAbs that have previously been reported to have broad activity. Although most of these antibodies target a similar site within the receptor binding domain (RBD) of the SARS-CoV-2 S protein and can be classified as class 4 antibodies (18), there is variability in the exact epitope and angle of approach. Consequently, we find that the exhibited breadth and potencies among these 12 mAbs greatly differ. One of the mAbs identified in this study, , was the only mAb to neutralize or bind all sarbecoviruses tested. We found that 10-40 and many other broadly neutralizing mAbs use a common motif in their heavy-chain complementarity-determining region 3 (CDRH3) gene, suggesting that it may be possible to elicit such antibodies in a general manner. Last, 10-40 exhibited in vivo efficacy against both SARS-CoV-2 and SARS-CoV, suggesting its use as a therapeutic agent. Collectively, we not only identify three broadly sarbecovirus neutralizing mAbs but also present a comparative evaluation of broadly neutralizing mAbs. The information provided herein could aid the development of pan-sarbecovirus antibodies and vaccines. the SARS-CoV S protein (Fig. 1A). All three antibodies recognized epitopes within the RBD (Fig. 1A) and inhibited the binding of soluble human angiotensin converting enzyme 2 (ACE2) receptor to the S protein (Fig. 1B). Epitope mapping by competition enzymelinked immunosorbent assay (ELISA) was carried out on these three antibodies, along with a panel of nine RBD-specific mAbs reported to have breadth against sarbecoviruses, including DH1047 (8), S2X259 (9), REGN10985 (10), ADG-2 (11), 2-36 (12), COVA1-16 (13), CR3022 (14), S2H97 (15), and S309, also known as sotrovimab (16). 10-40, 10-28, and 11-11 fell into one competition group with seven other mAbs ( Fig. 1C and fig. S3) that are known to recognize an inner face of RBD when it is in the "up" position (8)(9)(10)(11)(12)(13). The epitope of S2H97 was partially overlapping whereas that of S309 was discrete; this was not unexpected because the latter is directed to an epitope on the outer face on RBD (15,16). The binding affinities of this panel of mAbs to SARS-CoV-2 and SARS-CoV S proteins were also measured by surface plasmon resonance (SPR) ( fig. S4).
We next evaluated each mAb for neutralization against 10 non-SARS-CoV-2 sarbecoviruses capable of using human ACE2 as receptor in a VSVGpseudotyped virus assay (Figs. 2C and fig. S8). Only 10-40 and DH1047 neutralized all sarbecoviruses tested, whereas 11-11, S2X259, and ADG-2 were deficient or weak in neutralizing at least one of the viruses in the panel. The other mAbs had many more "holes" in their repertoire. As expected, none of these antibodies neutralized the Middle East respiratory syndrome virus (MERS), a member of the merbecovirus subgenus.
There are more sarbecoviruses found in bats in Africa and Europe or Asia ( fig. S9A) that do not use human ACE2 as receptor (17,20). Because their target cells are unknown, performing virus neutralization assay is not readily feasible. We therefore examined the binding profiles of this panel of mAbs to RBD proteins derived from six sarbecoviruses outside of SARS-CoV-2 and SARS-CoV sublineages ( Fig. 2D and fig. S9B). 10-40, 10-28, and S2H97 bound all RBDs tested, whereas DH1047 did not recognize the RBD of Rf1. The remaining mAbs bound only a subset of the RBDs. In particular, S2X259, REGN10985, ADG-2, and S309 did not recognize the RBD of most Asian bat sarbecoviruses. Similarly, we assessed the binding of this panel of mAbs to three S proteins derived from Asian bat sarbecoviruses ( fig. S9A) as expressed on the surface of transfected cells. 10-40, 10-28, 2-36, and DH1047 exhibited binding, but other mAbs were largely nonreactive ( Fig. 2E and fig. S10).
Given the recent rise of the Omicron variant of SARS-CoV-2, we placed a particular focus on how the many mutations found in the variant affected neutralization by the mAbs in our panel (Fig. 3A). Previous studies demonstrated that class 4 antibodies were considerably affected by the Omicron variant, B.1.1.529.1 (also known as BA.1), albeit in a disparate manner (4). We expanded on these previous findings by assessing our entire panel of 12 mAbs against B.1.1.529.1 (Fig. 3, B and C). All of the mAbs were adversely affected, with the exception of S2H97. Intriguingly, we found that DH1047, REGN10985, and ADG-2 lost more than 100-fold potency, whereas others, including 10-28 and S309, only lost a fewfold. 10-40 was moderately affected but could still achieve 100% neutralization at the highest dose. As these mAbs are not linked by a shared epitope according to their structures with the wild-type S protein, this suggests that the S protein of B.1.1.529.1 may have a slightly altered conformation in this particular region. In summary, these findings show that 10-40 is a broadly neutralizing RBD-directed mAb. Structural studies of 10-40, 10-28, and 11-11 To investigate the nature of antibody-S protein interactions for 10-40, 10-28, and 11-11, we determined the cryo-electron microscopy (cryo-EM) structures for antigen-binding fragments (Fabs) of these mAbs in complex with S2P prefusion-stabilized S proteins from SARS-CoV-2 WA1 and B.1.351 strains. We observed a greater degree of S protein disassembly for all WA-1 complexes, and higher-quality cryo-EM maps were obtained for all three B.1.351 complexes. A single predominant population was observed where three Fabs were bound per S protein in a 3-RBD-up conformation (Fig. 4A, figs. S11 to S13, and table S1). For 11-11, an additional class of two Fabs bound with 2-RBD up was also observed. The 10-40 complex reconstruction reached 3.5-Å global resolution, but local refinement of the RBD + Fab maps did not surpass 4-Å resolution. To resolve the interfaces, we determined crystal structures of the 10-40 and 10-28 Fab:RBD complexes. Both the crystal structures of 10-40 and 10-28 fitted nearly perfectly in the cryo-EM reconstruction density (figs. S11 and S12). For 11-11, a homology model was built and docked into the map ( fig. S13), which showed that this antibody recognizes RBD in the same way as S2X259 (9), with both sharing a similar light chain and CDRH3 motif.
To visualize the epitopes of 10-40, 10-28, and 11-11 on the SARS-CoV-2 WA1 RBD at higher resolution, we determined the structures of Fab in complex with RBD using x-ray crystallography, with suitable crystals obtained for 10-40 and 10-28 (table S2). The 10-40 crystal structure at 1.5-Å resolution revealed recognition of an epitope that is highly conserved among sarbecoviruses on the inner face of RBD (Fig. 4, B and C). 10-40 used three of its six CDR loops, H1, H3, and L2, to interact with an epitope consisting of a loop on RBD (residues 377 to 385) and extended toward the RBD ridge near the ACE2 binding site. 10-40 also established extensive polar contacts and hydrophobic interactions with RBD residues (Fig. 4B).
For the 10-28 crystal structure ( fig. S14A), antibody-RBD sidechain interactions were well defined at 3.2-Å resolution. Interactions were mediated by 10-28 CDR loops H1, H3, L1, and L3, which predominantly contacted not only 3 but also 2 and the 2-3, 4-5, and 5-4 loops of the RBD. The heavy chain formed three hydrogen bonds and a single salt bridge, between D95 of the CDRH3 and K386 of the RBD, whereas the light chain residues formed a total of five hydrogen bonds (fig. S14B).
All three antibodies recognized a region on the inner side of RBD that is hidden in the RBD-down conformation. Thus, they can

of 16
only recognize RBD in the up conformation. The 10-40 epitope is similar to the previously defined "class 4" antibody epitope, although we believe that all three of these antibodies can be classified as such (18). Superposition of the 10-40 Fab with the ACE2-RBD complex (6M0J) showed that antibody binding places the variable domain of the light chain (VL) of the antibody in a position that would clash with ACE2 (Fig. 4D), consistent with the experimental data showing inhibition of ACE2 binding (Fig. 1B).
Structural comparison of these newly identified antibodies with those previously identified targeting this class 4 epitope revealed that several modes of binding are available to this site (fig. S15) (8,9,(11)(12)(13)(14)(21)(22)(23). Similar antibodies were available for each of 10-40, 10-28, and 11-11. Although we observed a relationship between the epitope and the breadth exhibited by these antibodies, there also seems to be additional factors involved, such as the binding angle.
Given the binding breadth exhibited by 10-40 ( Fig. 2), we conducted additional structural studies for this mAb to investigate the basis for its cross-reactivity. To this end, we determined crystal structures of the 10-40 Fab bound to the RBDs of the sarbecoviruses SHC014, RaTG13, and WIV1 at 2.2, 2.8, and 3.1 Å resolutions, respectively (Fig. 5, A and B, and table S3). These structures

of 16
demonstrated that 10-40 bound to a highly conserved epitope that allowed for similar binding of 10-40 to these sarbecoviruses and SARS-CoV-2.
As observed in the complex of 10-40 bound to SARS-CoV-2 RBD, most of the interactions for 10-40 with SHC014, RaTG13, and WIV1 were through its CDRH3 and CDRL2 (Figs. 4B and 5A). We found that additional hydrogen bonds and salt bridges were present between 10-40 and these three sarbecoviruses as compared to SARS-CoV-2. Specifically, although the 10-40-SARS-CoV-2 interface involved 10 hydrogen bonds and three salt bridges, there were 15, 14, and 17 hydrogen bonds and four, three, and three salt bridges for 10-40 with SHC014, RaTG13, and WIV1, respectively (Fig. 5B). These additional contacts may explain why the potency for 10-40 against SHC014, RaTG13, and WIV1 is greater than that for SARS-CoV-2 ( Fig. 2).
Among the four complexes of 10-40 with sarbecovirus RBDs, there were a total of 12 interactions (nine hydrogen bonds and three salt bridges) that were conserved in all of the complexes (Fig. 5B). Such extensive overlap suggests that for mAbs that have deficiencies against some of these sarbecoviruses (Fig. 2), engineering of similar interactions could help to improve their breadth.

Identification of a common motif among broadly neutralizing mAbs
Given the broad recognition of 10-40 for sarbecoviruses, we carefully analyzed the amino acids that form its epitope and noted their remarkable conservation among sarbecoviruses ( Fig. 6A and fig.  S16), suggesting that this RBD region was likely subjected to strong functional constraints during evolution of this subgenus. Residues 377 to 379 in a -sheet were specifically targeted by the CDRH3 of 10-40 through multiple hydrogen bonds (Fig. 6B). COVA1-16, 2-36, and C022 interacted quite similarly with the same residues. We also noticed that these three mAbs, together with 10-40, contact this particular -sheet through a "YYDRSGY" motif originating from IGHD3-22 (Fig. 6B), a D gene that is frequently used by antibodies in the human repertoire (fig. S17). Within publicly available SARS-CoV-2 antibody sequences, this motif was found in 15 antibodies and was almost exclusively found in broad RBD-directed  (57), BD-801 is from (58), C022 is from (59), 2-36 is from (12), C634 and C652 are from (21), C850 and C996 are from (60), C1243, C1437, and C1332 are from (61), and C049 is from (62).
antibodies targeting this same epitope (Fig. 6C). This motif contains a Ser-to-Arg substitution, which formed hydrogen bonds with 369Y and 371S on RBD (Fig. 6A). We believe that these structural similarities define 10-40, 2-36, C022, and COVA1-16 as members of a new antibody class, each using a shared mode of heavy chain binding to RBDs of sarbecoviruses (Fig. 6C). As these four mAbs use diverse heavy chain V genes and light chain recombinations (table S4) S18B). We then performed a prevention experiment (Fig. 7A), administering 10-40 or an anti-HIV-1 control mAb 24 hours before the mice were challenged intranasally with MA10. Compared to the control group, significant weight loss was prevented (10 mg/kg, P = 0.7422) or mitigated (2 mg/kg, P = 0.0234) by 10-40 administration (Fig. 7B). In mice given the control antibody, high titers of infectious virus were observed in the lungs [greater than 10 5 median tissue culture infectious dose (TCID 50 )/g lung], whereas little (less than 10 4 TCID 50 /g lung) or no infectious virus was found in mice given 10-40 at 2 or 10 mg/kg, respectively (Fig. 7C). An analogous prevention experiment was conducted against SARS-CoV in hACE2transgenic mice (Fig. 7D). Weight loss was again prevented by 10-40 administration (Fig. 7E), and titers of SARS-CoV were markedly reduced in the lungs of mice pretreated with 10-40 (Fig. 7F). 10-40 appeared to be active in vivo against both SARS-CoV-2 and SARS-CoV.

DISCUSSION
Sarbecoviruses have caused two major outbreaks in humans in the past two decades: SARS-CoV in 2002 to 2003 and SARS-CoV-2 now. Thus, the world must prepare for the possibility of a future epidemic or pandemic due to another member of this subgenus that is presently harbored by bats and other animals (5,17,26). Pansarbecovirus neutralizing mAbs and vaccines could be useful interventions to contain another outbreak. Here, we have identified a human mAb, 10-40, that suits this need. 10-40 neutralized or bound to every sarbecovirus we tested, regardless of their usage of ACE2 as receptor. It exhibited potency against SARS-CoV-2 in vitro and in vivo, and its potency against other sarbecoviruses was even better, despite being isolated from a patient with COVID-19. Very recently, three sarbecoviruses closest genetically to SARS-CoV-2 have been identified in bats in Laos (26), along with two more in a different sublineage. Although not empirically tested, we note that they are likely to be susceptible to 10-40 because the key amino acids that would form this RBD epitope are identical to those in either GD-Pangolin or RmYN02, both of which were neutralized or bound by 10-40. Therefore, we believe that 10-40 is a promising candidate for pandemic preparedness. As SARS-CoV-2 continues to evolve, new variants may arise that elude current mAbs but remain susceptible to 10-40. It is imperative that we are prepared for a future zoonotic spillover event that may well occur.
The pursuit of a pan-sarbecovirus vaccine is already underway, including strategies that specifically target the stem helix in the S2 region of the S protein (27) or conserved elements within the RBD. Efforts directed to the latter have already shown promise (28)(29)(30). Our comparison of the antibodies directed to the conserved epitope delineated in this study highlight the structural differences in the epitopes recognized by 10-40 and DH1047 versus epitopes recognized by mAbs (such as ADG-2 or S2X259) with lower breadth against sarbecoviruses, which could be informative in focusing the antibody response to certain conserved residues on the inner face of RBD. The epitopes of 10-40 and DH1047 are substantially overlapping and target the same exposed -sheet, but their angles of approach are different, suggesting that this site could be attacked in more than one way. This angle may be critical as antibodies with similar affinities, such as 10-40, COVA1-16, and 2-36, have differing potencies, although this will need to be investigated. Moreover, by genetically comparing 10-40 with other antibodies, we have uncovered a unique class of mAbs that target this particular -sheet in RBD through a common CDRH3 motif. That this class of mAbs could use multiple heavy and light chain V genes and did not require extensive somatic hypermutation is certainly welcome news for the development of a pan-sarbecovirus vaccine targeting the RBD.
There are some limitations to this study to be considered when interpreting the results. Although we investigated many RBDs and S proteins for binding and neutralization studies, we could not exhaustively test all known sarbecoviruses. Furthermore, for the non-human ACE2 using sarbecoviruses, we could not assay for neutralization and only quantified antibody binding. For the identified CDRH3 motif, we did not experimentally validate whether it could be successfully used in vaccination. Additional studies are warranted to further investigate these possibilities. Nevertheless, the identification of 10-40 as a potential pandemic preparedness agent and the comparative analyses presented herein could help to guide future anti-coronavirus efforts.

Study design
The objective of this study was to identify broadly neutralizing antibodies against sarbecoviruses and then to conduct a thorough comparative analysis between these antibodies with those previously known. To do so, we screened serum samples from convalescent COVID-19 patients for those with broad neutralizing capability, sorted for S protein-binding memory B cells from these individuals, and then expressed and tested the binding and neutralization capability of the identified antibodies. The three that were identified to have broad reactivity, 10-40, 10-28, and 11-11, were then interrogated in relation to other known antibodies, which we determined from a literature search. Binding and neutralization experiments were conducted in technical duplicate or triplicate, respectively, and were experimentally replicated three times, of which a representative result is shown. For the most promising antibody, 10-40, we proceeded to solve structures by cryo-EM and x-ray crystallography and validate its efficacy in vivo against both SARS-CoV-2 and SARS-CoV. The mice in the in vivo experiments were allocated randomly with sample sizes similar to those used in related studies (n = 6 to 8 per group), which we have previously observed to produce replicable results. Experimenters were not blinded for the experiments in this study, and no data were excluded.

Sample collection
The protocols for acquisition of samples were reviewed and approved by the Institutional Review Boards of Columbia University (protocol no. IRB-AAAS9722) and Hackensack Meridian School of Medicine (protocol no. Pro2020-0633). Informed consent was obtained from both patients. Patient 10 was symptomatic for COVID-19 in March 2020, and blood collection was performed in April 2020. Patient 11 was symptomatic for COVID-19 in November 2020, followed by two doses of mRNA-1273 vaccine in January and February 2021, and blood collection was performed 1 week after the second vaccination. Sequencing confirmed that patient 11 was infected with the R.1 variant of SARS-CoV-2.

Expression and purification of S trimer and RBD proteins
The mammalian expression vector encoding the ectodomain of the SARS-CoV S trimer and the vectors encoding SARS-CoV and SARS-CoV-2 RBD fused to an HRV-3C protease cleavage site followed by a mFc tag and an 8× His tag at the C terminus were provided by J. McLellan (31). The ectodomains with 2P and furin mutations of SARS-CoV-2 D614G and B.1.351 S trimers were synthesized, fused to an 8× His tag at the C terminus, and then cloned into the paH vector. Other sarbecovirus RBD-His expression vectors were provided by P. J. Bjorkman (28). To purify the S trimer proteins and RBD proteins, each expression vector was transiently transfected into Expi293 cells using polyethylenimine (PEI) (1 mg/ ml; Polysciences). Five days after transfection, the SARS-CoV and SARS-CoV-2 RBDs were purified using protein A agarose (Thermo Fisher Scientific). To obtain untagged RBD of SARS-CoV and SARS-CoV-2, the mFc and 8× His tags at the C terminus were removed by HRV-3C protease (Millipore-Sigma) and then purified using nickelnitrilotriacetic acid (Ni-NTA) resin (Invitrogen), followed by protein A agarose. The S trimers and other sarbecovirus RBDs were purified using Ni-NTA resin (Invitrogen).

Identification of S trimer-specific antibody transcripts
We used an analysis protocol similar to our previous study to identify S protein-specific antibody transcripts (19). Briefly, full-length antibody transcripts were assembled using the Cell Ranger V(D)J pipeline (version 3.1.0, 10X Genomics) with default parameters and the GRCh38 V(D)J germline sequence version 2.0.0 as reference. To distinguish cells from the antigen sort and spike-in, we first used the count module in Cell Ranger to calculate copies of all hashtags in each cell from the Illumina next-generation sequencing raw reads. High-confidence antigen-specific cells were identified as follows. A cell must contain more than 100 copies of the antigen sort-specific hashtag to qualify as an antigen-specific cell. Because hashtags can fall off cells and bind to cells from a different population in the sample mixture, each cell usually has both sorted and spiked-in-specific hashtags. To enrich for true antigen-specific cells, the copy number of the specific hashtag has to be at least 1.5× higher than that of the nonspecific hashtag. Low-quality cells were identified and removed using the cell-calling algorithm in Cell Ranger. Cells that did not have productive heavy and light chain pairs were excluded. If a cell contained more than two heavy or light chain transcripts, then the transcripts with fewer than three unique molecular identifiers were removed. Cells with identical heavy and light chain sequences, which may be from mRNA contamination, were merged into one cell. We also applied additional filters to remove low-quality cells and transcripts in the antibody gene annotation process.

Antibody transcript annotation
Antigen-specific antibody transcripts were processed using our bioinformatics pipeline SONAR version 2.0 for quality control and annotation (32). Briefly, V(D)J genes were assigned for each transcript using BLASTn with customized parameters against a germline gene database obtained from the international ImMunoGeneTics information system (IMGT) database (33). On the basis of BLAST alignments of V and J regions, CDR3 was identified using the conserved second cysteine in the V region and WGXG (heavy chain) or FGXG (light chain) motifs in the J region (X represents any amino acid). For heavy chain transcripts, the constant domain 1 (CH1) sequences were used to assign isotype using BLASTn with default parameters against a database of human CH1 genes obtained from IMGT. A BLAST E value threshold of 10 −6 was used to find significant isotype assignments, and the CH1 allele with the lowest E value was used. Sequences other than the V(D)J region were removed, and transcripts containing incomplete V(D)J regions or frame shifts were excluded. We then aligned each of the remaining transcripts to the assigned germline V gene using CLUSTALO and calculated the degree of somatic hypermutation (34). The D gene assignment for each transcript was performed by the HighV-QUEST function in IMGT web server with the default parameters. For cells having multiple high-quality heavy or light chains, which may be from doublets, we synthesized all H and L chain combinations.

Antibody expression and purification
For each antibody, variable genes were optimized for human cell expression and synthesized by GenScript. VH and VL were inserted separately into pcDNA3.4 plasmids encoding the constant region for heavy chain and light chain, respectively. mAbs were expressed in Expi293F cells (Thermo Fisher Scientific, catalog no. A14527, RRID: CVCL_D615) by cotransfection of heavy chain-and light chainexpressing plasmids using PEI (Polysciences) and cultured in a 37°C shaker at 125 rpm under 8% CO 2 . Supernatants were collected 5 days after transfection, and then antibodies were purified by affinity chromatography using rProtein A Sepharose (Cytiva).

Antigen binding testing by ELISA
Fifty nanograms per well of antigen (S trimer or RBD protein) was coated onto ELISA plates at 4°C overnight. The ELISA plates were then blocked with 300 l of blocking buffer consisting of phosphate-buffered saline (PBS) with 1% bovine serum albumin and 20% bovine calf serum (Sigma-Aldrich) at 37°C for 2 hours. Afterward, 100 l of serially diluted antibodies was added and then incubated at 37°C for 1 hour. Next, 100 l of 10,000-fold diluted Peroxidase AffiniPure goat anti-human IgG Fc fragment-specific antibody (Jackson ImmunoResearch, catalog no. 109-035-170, RRID: AB_2810887) was added into each well and incubated for another 1 hour at 37°C. The plates were washed between each step with PBST (0.5% Tween-20 in PBS). Last, 3,3′,5,5′-tetramethylbenzidine (TMB) substrate (Sigma-Aldrich) was added and incubated before the reaction was stopped using 1 M sulfuric acid. Absorbance was measured at 450 nm.
Antigen binding testing by SPR SPR binding assays for IgG binding to S protein were performed using a Biacore T200 biosensor, equipped with a Series S CM5 chip (Cytiva), in a running buffer of 10 mM HEPES (pH 7.4), 150 mM NaCl, 3 mM EDTA, and 0.05% P-20 (HBS-EP+ buffer, Cytiva) at 25°C. SARS-CoV-2 or SARS-CoV S protein was captured through its C-terminal His-tag over an anti-His antibody surface. These surfaces were generated using a His-capture kit (Cytiva) according to the manufacturer's instructions, resulting in about 10,000 resonance units (RUs) of anti-His antibody over each surface. S protein was captured over a single flow cell at 125 to 200 RUs. An anti-His antibody surface was used as a reference flow cell to remove bulk shift changes from the binding signal. IgG antibodies were tested using a threefold dilution series of IgG antibodies with concentrations ranging from 1.2 to 33.3 nM. The association and dissociation rates were each monitored for 55 and 300 s respectively, at 50 ml/ min. The bound S protein-IgG complexes were regenerated from the anti-His antibody surface using 10 mM glycine (pH 1.5). Blank buffer cycles were performed by injecting running buffer instead of IgG to remove systematic noise from the binding signal. The resulting data were processed and fitted to a 1:1 binding model using Biacore Evaluation Software.

ACE2 competition and epitope mapping by ELISA
For the competition ELISA, purified antibodies and ACE2 protein (Sino Biological) were biotin-labeled using the One-Step Antibody Biotinylation Kit (Miltenyi Biotec) following the manufacturer's instructions and purified using a 40K molecular weight cutoff desalting column (Thermo Fisher Scientific). Serially diluted competitor antibodies (50 l) were added into S trimer-precoated ELISA plates, followed by 50 l of biotinylated antibodies at a concentration that achieved an OD 450 (optical density at 450 nm) reading of 2.5 in the absence of competitor antibodies. Plates were incubated at 37°C for 1 hour, and then 100 l of 500-fold diluted avidin-horseradish peroxidase (Thermo Fisher Scientific) was added into each well and incubated for another 1 hour at 37°C. The plates were washed with PBST between each of the steps. The plates were developed afterward with TMB, and absorbance was read at 450 nm after the reaction was stopped with 1 M sulfuric acid. For all the competition ELISA experiments, the relative binding of biotinylated antibodies and ACE2 to the S trimer in the presence of competitors was normalized by comparing to competitor-free controls. Relative binding curve and the area under curve (AUC) were determined in GraphPad Prism 9.2.

Cell surface S protein binding by flow cytometry
Expi293 cells were cotransfected with vectors encoding pRRL-cPPT-PGK-GFP (Addgene) and full-length S trimer of SARS-CoV, SARS-CoV-2, or other sarbecovirus at a ratio of 1:1. Three days after transfection, cells were incubated with antibody (10 g/ml) at 4°C for 1 hour. Then, 100 l of APC anti-human IgG Fc (BioLegend, catalog no. 366906, RRID: AB_2888847) at 1:20 dilution was added to the cells and incubated at 4°C for 45 min. Cells were washed three times with fluorescence-activated cell sorting buffer (PBS + 2% FBS) before each step. Last, cells were resuspended and antibody binding to cell surface S trimer was quantified on a LSRII flow cytometer (BD Biosciences). The mean fluorescence intensity of antibodybound APC-positive cells within green fluorescent protein (GFP)positive cells was analyzed using FlowJo (BD Biosciences).

Production of pseudoviruses
Recombinant VSV pseudoviruses in which the native glycoprotein was replaced with SARS-CoV S protein, S protein from SARS-CoV-2 or one of its variants, or other sarbecovirus S proteins were generated as previously described (36). Human embryonic kidney (HEK) 293T cells [American Type Culture Collection (ATCC), catalog no. CRL-3216, RRID: CVCL_0063] at a confluency of 80% were transfected with a S protein expression vector using PEI (1 mg/ml) and cultured overnight at 37°C under 5% CO 2 . Twenty-four hours later, cells were infected with VSV-G-pseudotyped G-luciferase (G*G-luciferase, Kerafast) at a multiplicity of infection (MOI) of 3 for 2 hours. Afterward, cells were washed three times with 1× PBS, changed to fresh medium, and cultured at 37°C for another 24 hours before supernatants were harvested and clarified by centrifugation at 300g for 10 min.

Pseudovirus neutralization assay
Pseudoviruses were first titrated to equilibrate viral input between assays. Neutralization assays were then performed by incubating pseudoviruses with serially diluted heat-inactivated serum or antibodies in triplicate in a 96-well plate for 1 hour at 37°C. For neutralization of SARS-CoV and SARS-CoV-2 pseudoviruses, Vero-E6 cells (ATCC, catalog no. CRL-1586, RRID: CVCL_0574) were seeded at a density of 4 × 10 4 cells per well, whereas for neutralization of pseudoviruses derived from other sarbecoviruses, 293T-hACE2 cells (36) were seeded at a density of 1 × 10 5 cells per well. Then, luciferase activity was measured using the Luciferase Assay System (Promega), according to the manufacturer's instructions, 24 hours after cells were added to the pseudovirus and antibody or serum mixture. The neutralization curves and IC 50 values were generated by fitting a nonlinear five-parameter dose-response curve in GraphPad Prism 9.2. Virus infectious titer was determined by an end-point dilution and cytopathic effect (CPE) assay on Vero-E6 cells as described previously (19).

Neutralization of authentic SARS-CoV-2 by purified mAbs
An end-point dilution assay in a 96-well plate format was performed to measure the neutralization of SARS-CoV-2. Briefly, purified mAbs were serially diluted at fivefold dilutions starting at 50 g/ml. Triplicates of each dilution of the antibody were incubated with SARS-CoV-2 at an MOI of 0.1 in Eagle's minimum essential medium (ATCC) with 7.5% inactivated FBS for 1 hour at 37°C. After incubation, the antibody-virus mixture was transferred onto a monolayer of Vero-E6 cells grown overnight. The cells were incubated with the mixture for 70 hours. Morphological changes resulting from CPE due to infection of cells were visually scored for each well from 0 to 4 with 4 defined as the appearance of complete virus-mediated cytopathy. Double-blinded scoring of the CPE was converted to percentage of neutralization, and the IC 50 was determined by fitting a five-parameter dose-response curve using GraphPad Prism v9.2.

Neutralization of authentic SARS-CoV by purified mAbs
To measure neutralization of authentic SARS-CoV, Vero-E6 cells were seeded in 96-well plates in cell culture media [Dulbecco's modified Eagle's medium (DMEM) + 10% FBS + 1% P/S] overnight at 37°C under 5% CO 2 to establish a monolayer. The next day, antibodies were diluted as appropriate in DMEM + 2% FBS and incubated with 0.01 MOI of SARS-CoV GZ50 strain (GenBank accession no. AY304495) for 1 hour at 37°C in quadruplicate (37). After incubation, the antibody-virus mixture was overlaid onto cells and incubated at 37°C under 5% CO 2 for 72 hours. CPE was then visually scored as either negative or positive for infection by comparison to control uninfected or infected wells in a blinded manner by two independent observers. Each concentration of antibody was tested in triplicate, and the percentage of infected wells was used to calculate IC 50 by fitting a five-parameter dose-response curve using GraphPad Prism v9.2.

Protein expression for cryo-EM and crystallography
The RBD (residues 319 to 541) of the SARS-CoV-2 was expressed and purified as previously described (38). Briefly, protein expression was carried out in suspension cultures of HEK 293 GnTI-Freestyle cells using serum-free media (Invitrogen) by transient transfection using PEI (Polysciences). The medium was harvested 4 days after transfection, and the secreted protein was purified from the supernatant using nickel affinity chromatography using Ni-NTA IMAC Sepharose 6 Fast Flow resin (GE Healthcare) followed by size exclusion chromatography (SEC) on a Superdex 200 column (GE Healthcare) in 20 mM tris and 150 mM NaCl (pH 8.0).
Fab fragments of 10-40, 10-28, and 11-11 were produced by digestion of IgG antibodies with immobilized Endoproteinase Lys-C (Sigma-Aldrich) equilibrated with 25 mM tris (pH 8.5) and 1 mM EDTA for 3 hours. The resulting Fabs were purified from the cleaved Fc domain by affinity chromatography using protein A. Fab purity was analyzed by SDS-polyacrylamide gel electrophoresis. All Fabs were buffer-exchanged into 20 mM tris and 150 mM (pH 7.4) before cryo-EM or crystallization experiments.

Cryo-EM grid preparation
Samples for cryo-EM grid preparation were produced by mixing purified B.1.351 S protein (final protein concentration of 1 mg/ml) with Fabs in a 1:3 molar ratio, followed by incubation on ice for 1 hour. The final buffer for the S protein + Fab complex was 10 mM sodium acetate and 150 mM NaCl (pH 5.5). A final concentration of 0.005% (w/v) n-dodecyl -d-maltoside was added to the mixtures to prevent preferred orientation and aggregation during vitrification. Cryo-EM grids were prepared by applying 3 l of the sample to a freshly glow-discharged carbon-coated copper grid (C-flat 1.2/1.3 300 mesh); the sample was vitrified in liquid ethane using a Vitrobot Mark IV with a wait time of 30 s, a blot time of 3 s, and a blot force of 0.

Cryo-EM data collection and analysis
Cryo-EM data for single-particle analysis was collected on a Titan Krios electron microscope operating at 300 kV, equipped with a Gatan K3-BioQuantum direct detection detector and energy filter, using the SerialEM software package (39). Exposures were taken with a total electron fluence of 41.93 e−/Å 2 fractionated over 48 frames, with a frame time of 50 ms and a total exposure time of 2.4 s. A defocus range of −0.8 to −2.0 m was used with a magnification of ×81,000 and a pixel size of 1.07 Å.
Data processing was performed using cryoSPARC v3.2.0 (40). Raw movies were aligned and dose-weighted using patch motion correction, and the contrast transfer function (CTF) was estimated using patch CTF estimation. Micrographs were picked using blob picker, and a particle set was selected using two-dimensional (2D) and 3D classification. The resulting particle set was local motion corrected and refined to high resolution using a combination of heterogeneous and homogeneous refinement. The final maps were deposited to the Electron Microscopy Data Bank (EMDB) with ID nos. EMD-25146 (10-40), EMD-25166 (10-28), and EMD-25167 .

Sequence conservation and phylogenetic analysis for sarbecoviruses
S protein sequences of 52 sarbecoviruses were obtained from the National Center for Biotechnology Information with the accession numbers listed in table S5. The sequences were aligned with the MUltiple Sequence Comparison by Log-Expectation algorithm in MEGA7 with the default parameter (52,53). The conservation of each RBD residue was calculated using the entropy function of the R package bio3d (H.norm column) based on the RBD sequence alignment. Sequence entropy was visualized on the RBD structure using PyMOL version 2.3.2. The maximum likelihood phylogenetic tree of sarbecoviruses was generated using MEGA7 by using the amino acid sequence alignment of RBDs with the general time reversible (GTR + G) substitution model, and the ancestor branch of BM4831 and BtY72 was used to root the phylogenic tree.

Epitope and paratope analysis for antibody 10-40
The paratope and epitope residues for 10-40 were identified using PISA with the default parameters. 10-40 epitope residues from different sarbecoviruses were identified from the RBD sequence alignment described above. Positional frequency of amino acids within the 10-40 epitope was generated using WEBLOGO (54). The genespecific substitution profiles for 10-40 germline genes and IGHD3-22 gene usage in antibody repertoires were obtained from the curated antibody repertoires (cAb-Rep) database (55).

Protection against mouse-adapted SARS-CoV-2 challenge by 10-40
The mouse study was performed at the Columbia University Medical Center under a protocol (AC-AABK7562) approved by the Institutional Animal Care and Use Committee. All work was performed with approved standard operating procedures, and the SARS-CoV-2 challenge was conducted in an animal biosafety level 3 facility. Sixteen-week-old female BALB/c mice (The Jackson Laboratory) were intraperitoneally injected with the indicated concentration of antibody 24 hours before infection. For infection, each mouse was anesthetized with a mixture of ketamine/xylazine and infected intranasally with 50 l of PBS containing 5 × 10 4 plaque-forming units (PFUs) of mouse-adapted SARS-CoV-2 MA10 strain [obtained from BEI Resources, NIAID, NIH (24)] (n = 8 per group). Mice were monitored for body weight changes at day 0 and day 3. At day 3 after infection, mice were euthanized, and lung tissue was harvested and homogenized for viral titer analysis. Lung viral load titer was determined by an end-point dilution assay.

Protection against SARS-CoV challenge by 10-40
The in vivo anti-SARS-CoV activity of 10-40 was evaluated in the established hACE2-transgenic (human ACE2-transgenic) mouse model as described previously (25). These experiments were conducted under a protocol (5862-21) approved by the Hong Kong University (HKU) Committee on the Use of Live Animals in Teaching and Research and complied with the Animal Research: Reporting of In Vivo Experiments (ARRIVE) guidelines. Briefly, male hACE2transgenic mice, aged 8 to 10 weeks, were obtained from the HKU Centre for Comparative Medicine Research and divided into different groups and intraperitoneally injected with the indicated concentration of antibody 24 hours before infection. The following day, mice were anesthetized with ketamine (100 mg/kg) and xylazine (10 mg/kg) and then intranasally inoculated with 20 l of PBS containing 2000 PFUs of SARS-CoV GZ50 strain (n = 6 per group). Mice were monitored for body weight changes each day. The animals were euthanized at 5 days post-infection (dpi), and lung tissue was harvested and homogenized for viral load quantitation by quantitative reverse transcription polymerase chain reaction as previously described (37).

Statistics
All raw, individual-level data are presented in data file S1. P values between groups were determined by two-tailed t test with Wilcoxon matched-pairing (Fig. 7, B and E) or two-tailed Student's unpaired t test (Fig. 7, C and F), using GraphPad Prism v9.2. All conducted statistical tests were two-tailed. P values of less than 0.05 were considered statistically significant. Data are presented as means ± SEM or SD as described in the corresponding figure legends. Curve fitting was conducted by fitting a nonlinear five-parameter doseresponse curve using GraphPad Prism v9.2.